Investigating soil moisture dynamics for improved applications of radar satellites in agricultural water management

Efficient land and water management strategies are essential in order to meet increasing crop productivity and resource demands in modern agriculture. Soil moisture plays a central role in determining efficient strategies because it provides information crucial for the selection of optimal tillage practices and for determining irrigation schedules. Radar satellites have been a source of soil moisture information at different spatio-temporal scales. Although they provide information for almost all weather and environmental conditions, they only measure soil moisture at the upper surface layer. This can be inadequate since most agricultural applications require soil moisture information over deeper layers (e.g. over rooting depth). This thesis therefore investigates the utility of radar satellites, particularly Sentinel-1, for agricultural water management by investigating surface and subsurface soil moisture dynamics. Focus is given on the direct applicability of Sentinel-1-derived surface soil moisture to estimate indicators suitable for agricultural land and water management. The main hypothesis in this research project is that the known sensitivity of Sentinel-1 to surface soil moisture can be further exploited to gain insights and estimates of subsurface conditions, which are valuable for deriving indicators relevant for agricultural water management.

The vertical variability between surface and subsurface soil moisture was investigated from time series datasets using statistical methods to examine coupling between the two. The identification of (de)coupled conditions determine whether surface soil moisture can directly represent subsurface conditions. Low residuals variance and strong lagged dependence were the criteria used to identify coupled surface - subsurface conditions. It is highlighted that the occurrence of decoupled conditions is not confined to dry periods, as commonly encountered in literature. Subsurface soil moisture dynamics, particularly the occurrence of preferential flow paths at measurement locations were inferred to have influenced decoupling during wetter surface conditions.

Subsurface soil moisture variability in a cultivated field was investigated to determine the impact of soil structure and vegetation activity on soil moisture dynamics, based on almost a year of situ soil moisture measurements. The results of the temporal stability analysis demonstrate that the contribution of vegetation to soil moisture variability is larger compared to soil structure. Furthermore, the impact of land management practices for different crop types can result in contrasting subsurface soil moisture contents within the same field. Temporally dynamic soil hydraulic properties were estimated from inverse modeling using in situ measurements. However, the changes in soil hydraulic properties may be challenging to determine since using a single type of flow regime may be insufficient.

The potential of the vegetation backscatter from Sentinel-1 to reflect root zone soil moisture conditions, especially during water-limited periods, was examined for three growing seasons (2016 – 2018). A decrease in the total backscatter measured by Sentinel-1 was already observed during the 2018 European summer drought. Further investigation of the backscatter components using the Water Cloud model reveals that the soil backscatter correlates well with root zone soil moisture for non water-limited conditions. However, water-limited conditions highlights the ability of the vegetation backscatter to reflect root zone conditions based on good correlations obtained between the two. Unlike the soil backscatter which has been extensively studied in the past, the vegetation backscatter is deemed as an untapped source that can potentially allow direct estimation of root zone soil moisture from radar satellites.

During saturated or near-saturated conditions, assessment of field trafficability is important to ensure good vehicular mobility and to mitigate the present soil compaction rates in agricultural fields. Soil moisture coupling between the surface and topsoil layer was found to coincide with tillage periods, and facilitates the direct use of Sentinel-1 for estimating field trafficability. Additionally, field measurements of soil strength over the top soil were related with surface soil moisture to generate a probabilistic measure of trafficability. Conditions favorable for traffic were found in early spring, and the changes in trafficability can be closely monitored because of the high temporal frequency of Sentinel-1.

The dynamic and non-linear behavior of soil moisture over time complicates the estimation of root zone soil moisture directly from surface soil moisture or using simple empirical relations, especially for decoupled conditions. A data-driven machine learning approach using Random Forest is applied to estimate daily root zone soil moisture from time series datasets. Similar to a process-based hydrological model combined with data assimilation, Random Forest achieves high, or even in slightly higher, accuracies for interpolation and comparable accuracies with for extrapolation of root zone soil moisture. Based on model residuals, however, estimates for extreme dry and wet conditions are less accurate than commonly encountered intermediate soil moisture states. It is inferred that poor learning of Random Forest for such infrequently encountered extreme conditions and limitations of the pore-flow model applied for the process-based model contributed to the said findings. One of the advantages of RF is it does not make assumptions on the system dynamics, implying that information on soil hydraulic properties are not required for interpolation and extrapolation.

Overall, the results in this PhD research provide novel and innovative methods for applying Sentinel-1 that can collectively be incorporated into an adaptive framework that supports operational water management in agriculture. For saturated conditions at the beginning of the growing season when soil compaction is a concern, field trafficability can be directly estimated from Sentinel-1 derived surface soil moisture, which can assist farmers in determining the onset of tillage activities. Further into the growing season when adequate soil water supply is essential to ensure sufficient crop productivity, information on root zone soil moisture could potentially be derived from the vegetation backscatter especially for droughts or by integrating Sentinel-1 surface soil moisture into data-driven methods. These methods may be beneficial for data-poor regions with limited information on soil hydraulic properties. It is further highlighted that soil properties in agricultural areas are constantly changing and require dynamic soil functions and model structures in order for improved representation of subsurface processes that are essential for accurate soil moisture estimation.